Why the four ingredients matter

CAPE is gasoline. It measures how much energy a rising parcel of air has available to it. High CAPE means strong, deep updrafts that can reach 60+ mph upward — strong enough to suspend hailstones, strong enough to keep precipitation out of the rotation column.

Shear is the twist. If the wind direction and speed change with height — say, south at 15 mph at the ground, southwest at 35 mph at 3,000 ft, west at 60 mph at 6,000 ft — that creates a horizontal rolling motion in the air column. The updraft tilts that horizontal roll into the vertical, and the storm starts to rotate. Without shear you get a thunderstorm; with shear you get a supercell.

Moisture is the trigger condition. Without enough water vapor near the surface, rising air doesn't condense enough to release the latent heat that powers a deep convective cloud. Gulf of Mexico moisture is why the Southeast is a tornado factory in spring.

Lift is what gets the air parcel moving up in the first place — a cold front plowing in, a dry line, a warm front, an outflow boundary from a previous storm. The lift initiates; the other three ingredients decide what kind of storm you get.

Formation stage

What you're watching

Stage 1 · Wall cloud. A lowered, rotating base of cloud hanging from the southwest flank of the supercell. The mesocyclone — 2 to 6 miles wide — is the rotation aloft that the wall cloud is the visible bottom of. Most wall clouds don't produce tornadoes; the ones that do typically take 10 to 20 minutes to spin up first.

Stage 2 · Funnel cloud. Condensation extends downward from the wall cloud as the pressure inside the rotating column drops below the surrounding air. The drop in pressure cools the air enough to condense moisture — that's why you can see the funnel. The wind is rotating all the way to the ground long before the visible funnel reaches it.

Stage 3 · Touchdown. The pressure drop reaches the surface; if there's enough moisture, the funnel cloud becomes continuous to the ground. If the air is dry, you'll see a debris cloud at the base before the condensation funnel ever fills in.

Stage 4 · Mature. Full intensity. The funnel is widest, the debris cloud is largest, and the inflow at the base is moving 100+ mph horizontally toward the tornado.

Stage 5 · Rope-out. The tornado tilts, narrows, and contorts into a thin rope as the rear-flank downdraft cuts off the inflow. This is often the most photogenic and most dangerous stage — rope tornadoes can suddenly snap and change direction.

The math you're playing with

Two objects, two velocity vectors. The classic geometry problem is: given object A moving with velocity v_A from position p_A, and object B with v_B from p_B, when (if ever) are they closest, and how close do they get?

The answer is just algebra — you find the time when the squared distance is minimized, then check whether that distance is less than the radius of the storm's damage path. If your closest approach is less than the tornado radius, you got hit. If it's bigger, you punched through.

But the storm doesn't go in a straight line. Every minute or so it wobbles a few degrees — the rotation isn't perfectly steady, the surrounding flow shifts, the storm interacts with terrain. The uncertainty cone shows where the storm might be in 10 minutes, given how much it's wobbling. Run the simulation once and you get one outcome. Run it 25 times with the same dials and you get a probability — that's why the Monte Carlo button is there.

v0.1 note Two things had to change to make storm width matter. First: the radius slider now goes up to El Reno's 2.6-mile damage path, because the geometry can produce closest-approach values bigger than 1 mile and you were stuck capped below them. Second: the wobble was too tight — sigma had almost no visible effect on path variance, so every Monte Carlo run came out nearly identical and width either always-hit or never-hit.

v0.3.2 MATH FIX Then Travis noticed: at vertical (0°) and horizontal (90°) bearings the Monte Carlo fan was symmetric, but every diagonal bearing (45°, 135°, etc.) skewed the trails toward the "south end" of the spread. That's a real second-order curvature bias in the original "add a random angle to the bearing" model — Taylor-expand sin(b+w) and -cos(b+w) and the w² terms break left/right symmetry at diagonals. Fix: wobble the perpendicular velocity instead of the bearing angle (same sigma input, same OU damping, no curvature bias). Now the Monte Carlo fan is centered exactly on the deterministic trajectory at every bearing.

v0.2 BUG FIX The storm's starting position used to be anchored to the car's starting waypoint — 35 miles SW of wherever you began. That meant Birmingham start → storm at SW Birmingham; Tupelo start → storm at SW Tupelo. The corridor never crossed the storm path; the car was always ahead of the storm by design. v0.2 fixes this: the storm has a fixed origin regardless of where you put the car. Your car-start choice is a real choice about WHERE you are along I-22 when the storm crosses it.

v0.3 BUG FIX v0.2 picked a storm anchor that was actually off-canvas at (1.04, 0.96), and worse: the picture-rendering function still used the OLD car-derived anchor while the simulation used the new one. Picture and math diverged — that's why you saw the storm icon sitting south of Winfield but Monte Carlo reported 288,000-yard closest approach and the red trail lines disappeared. v0.3 unifies the anchor (now 0.41, 0.59, just SW of Winfield in real-world Marion County AL).

v0.3.3 — Why the default starts you in the path The defaults ship as an intercept geometry on purpose: car starting at Birmingham, heading NW toward Tupelo at 70 mph; storm bearing 68° NE at 29 mph out of Marion County, σ=8°, damage radius 2,500 yd. The storm's centerline crosses the corridor near mile 75 at t≈64 min — about when a 70-mph car reaches that mile mark. Run it 25 times and the car is inside the damage path in most of them.

That default is the lesson, and it is not a score. The number it produces is the one a real driver has no way to compute in the moment: if you keep driving toward where the storm is going, you do not get a coin flip — you get worse than a coin flip. The lab is set up to lose so that the alternatives cost you something to find. Switch to PULLED OVER and the crossing happens on an empty stretch of road. Switch to SE and you are driving out of the geometry instead of into it. Push σ to 15° and watch the certainty fall apart — because in the real thing, σ is never 8°, and nobody tells you what it is.

The point is the one the NWS has made for years, and this lab has no business making it any other way: you cannot outrun a tornado in a car. The winning move here is the boring one. It is at the top of this page, in the shelter guidance, and it does not photograph well. Ninety-five people died along the two tracks this lab borrows its map from. The simulation is a toy; the geometry is not; and the reason the toy is worth running is that it makes the boring answer feel earned instead of preached.